Amog Kamsetty

@sergii-ivakhno-kidsloop could you share the full stack trace? I think you would just need to remove the `gpus=-1` arg from the `Trainer` since you already have `use_gpu=True` set in the...

And also can you try with adding `ray.init("auto")` at the beginning of your script.

also running into this error with 3.5.1 when trying to use stage level scheduling with `dynamicAllocation` set to false ``` : org.apache.spark.SparkException: ResourceProfiles are only supported on YARN and Kubernetes...

the exception being thrown is still in master https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/resource/ResourceProfileManager.scala#L90-L91

Thanks for reporting the issue @n30111! Indeed, definitely something we should fix. I think we should switch to using the Tune Session API internally instead of `tune.checkpoint_dir`, and then on...